List of Papers Browse by Subject Areas Author List
Abstract
Alzheimer’s Disease (AD) remains a major diagnostic challenge due to the complex interplay of genomic, radiomic, and structural factors in disease progression. While deep learning methods can classify AD, current approaches fail to effectively combine multimodal data with clinical knowledge, compromising both accuracy and interpretability. We present ClinGRAD, a clinically-guided heterogeneous graph neural network that combines genomic and radiomic data using connections based on diffusion-weighted imaging (DWI) maps and gene co-expression networks. ClinGRAD’s contributions include: (1) a multimodal fusion architecture that integrates validated structural and genetic connectivity patterns for consistent biological feature analysis; (2) a multi-scale graph framework capturing both local brain structure and global genomic pathway relationships; (3) an attention mechanism that provides clinically relevant explanations of gene-structure interactions; and (4) pathway-based gene clustering that reveals underlying biological mechanisms and their clinical implications. ClinGRAD outperforms existing models, achieving an accuracy of 93.15%, distinguishing AD from control, mild cognitive impaired, and vascular dementia patients while maintaining biological coherence through its clinical guidance framework. The code will be available at https://github.com/BioMedIA-MBZUAI/ClinGRAD.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/5205_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/BioMedIA-MBZUAI/ClinGRAD
Link to the Dataset(s)
N/A
BibTex
@InProceedings{HasSal_ClinGRAD_MICCAI2025,
author = { Hassan, Salma and Salem, Mostafa and Papineni, Vijay Ram Kumar and Elsayed, Ayman and Yaqub, Mohammad},
title = { { ClinGRAD: Clinically-Guided Genomics and Radiomics Interpretable GNN for Dementia Diagnosis } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15971},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
Authors propose ClinGRAD, a clinically-guided, interpretable heterogeneous graph neural network (GNN) architecture for dementia diagnosis, with a focus on Alzheimer’s Disease (AD). ClinGRAD integrates genomic and radiomic data using clinically validated structural and genetic connectivity information, including diffusion-weighted imaging (DWI) maps and gene co-expression networks. Its key innovations include a multimodal fusion strategy informed by biological priors, a multi-scale graph structure spanning molecular to anatomical domains, and an attention-based mechanism that provides interpretability by highlighting critical gene-structure interactions. Evaluated on the ANMerge dataset, ClinGRAD outperforms existing models—achieving up to 98.75% accuracy—and offers transparent diagnostic insights by identifying influential genes and brain regions involved in AD.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Clinically-Informed Multimodal Integration: The paper introduces a heterogeneous graph neural network (GNN) that incorporates clinically validated priors such as DWI-based anatomical connectivity and gene co-expression networks.
- Interpretability with Biological Relevance: ClinGRAD uses attention mechanisms and gene-pathway clustering to trace back predictions to influential genes and brain structures.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- Insufficient Explanation of the Core Method: Although the model relies on GAT-based heterogeneous message passing, the paper lacks a clear and detailed explanation of how clinical priors (e.g., DWI and gene pathways) influence the attention mechanism or the message-passing dynamics.
- Lack of Quantitative Interpretability Validation: While the model includes attention-based interpretability and gene clustering visualizations, it does not evaluate the interpretability quantitatively (e.g., alignment with expert-labeled biomarkers or clinical annotations).
- Lack of Clarity in Figure Presentation and Referencing: Figures 2 to 4, which depict gene clustering, brain region influence, and cross-pathway gene interactions, are not clearly introduced or contextualized within the Results or Discussion sections. Notably, Figure 4 is not explicitly referenced in the main body of the text, and its relevance to the narrative is unclear.
- Please rate the clarity and organization of this paper
Poor
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper lacks sufficient methodological detail to understand or verify the proposed approach, and it provides no information to support reproducibility, such as code, data access, or implementation specifics. These omissions significantly undermine the credibility and utility of the work.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
In this manuscript, the authors introduce ClinGRAD, a clinically guided heterogeneous graph neural network that fuses MRI derived radiomics features with gene expression profiles for dementia classification. The graph is constrained by diffusion weighted MRI (DWI) tractography and gene co expression priors, and an attention mechanism yields node and edge level explanations. On the ANMerge cohort, ClinGRAD outperforms eight uni and multimodal baselines and ablation variants.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Anatomically validated DWI edges and gene co expression weights are hard wired into the graph, enabling biologically plausible message passing.
- Separate patient, structure and gene nodes coupled through relation specific attention provide multi scale context and yield clear saliency maps that link influential genes to affected brain regions.
- The experimental design is decent with stratified 3 fold cross validation, comparisons with eight baselines (CNN, ViT, Flex MOE, FGCNN, FT Transformer, GCN, etc.) and systematic ablations of each modality and edge type.
- Helpful downstream analysis regarding pathway analysis.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- All results are confined to ANMerge; I am still not sure why the authors don’t evaluate using ADNI as it does contain genomic data.
- Only 75 pre selected AD genes are used. This is very narrow and does not reflect real-world settings. There is now explanation on the gene selection which could also introduce considerable bias.
- Figure 2 is difficult to read. Please increase the font size.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Method is interesting and shows strong performance. Data preprocessing is missing.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
The authors addressed the comments in their rebuttal
Review #3
- Please describe the contribution of the paper
The paper proposes a clinically-guided heterogeneous GNN, called ClinGRAD, which effectively integrates multimodal data with clinical knowledge for diagnosing Alzheimer’s disease (AD). By combining genomic and radiomic data, the model achieved superior performance in AD classification tasks, while its interpretability provides valuable insights into gene-gene interactions and brain regions affected by AD. It improves the clinical applicability and enables a more comprehensive understanding of the disease.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The model utilizes a heterogeneous GNN, which is powerful to capturing the complex biological relationships of data. In addition, its interpretability provides clear insights into gene-gene interactions and the brain regions most affected by AD, which enhances both its clinical applicability and our understanding of AD pathogenesis.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The model was only tested on just one dataset, and the size of data is not clear.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Due to the availability of a dataset that combines longitudinal MRI scans with comprehensive genomic data for dementia analysis, the model was tested on only one dataset. However, the experimental results have demonstrated the power and effectiveness of the GNN on this dataset, showing promising potential.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
The authors have addressed my concerns. I think the paper is ready to be accepted.
Author Feedback
We thank all reviewers (R1–R3) for their thoughtful evaluations and constructive feedback. We are pleased that the reviewers acknowledged the novelty of ClinGRAD’s clinically-informed heterogeneous GNN design, its interpretability through biologically grounded attention mechanisms, and its strong performance on multimodal dementia diagnosis (R1–R3). We also appreciate the recognition of its clinical relevance and methodological contributions (R2, R3).
R1 Response We thank Reviewer 1 for their comments; however, to clarify, our methodology extensively integrates clinical priors: DWI connectivity and gene co-expression scores are not merely used to define edge existence; they are embedded as explicit edge attributes that shape the attention-based message passing in our GAT framework. Specifically, for structure-to-structure edges, we compute edge weights using a learnt representation over both Euclidean distances and DWI-derived connectivity strengths, enforcing spatial and anatomical fidelity during aggregation. For gene-gene edges, we use co-expression scores to preserve functional gene relationships. These priors dynamically modulate the attention coefficients during graph updates, ensuring that message passing respects known biological structure instead of relying solely on learned correlations. This biologically constrained attention mechanism is a core innovation of our model and contributes significantly to both its accuracy and interpretability, as demonstrated in Table 2. Regarding interpretability, our visualizations in Figures 2-3 align with established AD literature [1] but we acknowledge the benefit of incorporating quantitative validation metrics, and we will include that in future work.
R2, R3 – Dataset scope and gene selection We thank the reviewers for this point. ANMerge is the only public dataset with longitudinal MRI, blood-based gene expression, and multi-class diagnostic labels for 1,700 participants. In contrast, ADNI provides whole-genome sequencing (WGS) for a subset of patients but lacks gene expression data, capturing only genomic variants rather than gene activity. Unlike WGS, which reflects static genetic variants, ANMerge’s gene expression data captures dynamic, tissue-specific activity, providing insight into which genes are actively driving disease, rather than just indicating genetic potential.
Regarding gene selection [R3], we selected 75 AD-associated genes representing the intersection between validated genes from the Human Protein Atlas [5] and genes available in the dataset, as described in Section 2. This approach minimizes bias by using externally validated AD biomarkers while ensuring data availability. We acknowledge expanding the gene set is a valuable next step and will explore this in future work, but we were limited in terms of publicly available data.
R1, R3 – Figure clarity and references We will revise the manuscript to clearly reference Figures 2–4 and improve their integration into the Results section.
R1, R2, R3 – Preprocessing, Reproducibility, and code availability
[R3] Preprocessing details, including segmentation, Radiomics feature extraction, and construction of graph edges, are summarized in Section 2.To maintain anonymity, we could not include a link to our code. Upon acceptance, we will release the full implementation including hyperparameters, preprocessing scripts, and trained model checkpoints. The dataset used is already publicly available.
R1 – Writing clarity and organization We acknowledge the concern about figure referencing. While R2 and R3 found the paper satisfactory, we will revise the structure to better integrate figures and improve the flow.
[1] Mahoney, E.R., et al.: Brain expression of the vascular endothelial growth factor gene family in cognitive aging and Alzheimer’s disease. Molecular psychiatry 26(3), 888–896 (2021).
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
I agree with the reviewer that this paper lacks sufficient methodological detail to fully understand or verify the proposed approach and does not provide the necessary information to support reproducibility.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A